Fertility Limits on Local Politicians in India∗

S Anukriti† Abhishek Chakravarty‡

September 19, 2014

Abstract

We examine the demographic implications of fertility limits on local politicians. Several

Indian states disbar individuals with more than two children from contesting Panchayat

and municipal elections. These two-child limits are intended to decrease fertility among

the constituents through a role-model effect and by incentivizing individuals who intend

to run for elections in the future to plan smaller families. We find that fertility limits

on elected representatives decrease voters’ likelihood of having more than two children.

However, they also increase the sex ratio of second births for politically dominant

upper-caste families with historically stronger preference for sons. Our results point

towards a novel source of demographic influence: political leaders.

Keywords: India, Local Elections, Fertility Limits, Sex Ratios, Population Control

∗We thank Heather Banic for excellent research assistance.†Department of Economics, Boston College. anukriti@bc.edu.‡Department of Economics, University of Essex. achakrb@essex.ac.uk.

0

1 IntroductionDeveloping countries with large populations often employ policy tools to decrease fertility.

Recent measures include direct fertility limits on citizens (e.g., China’s One Child Policy),

conditional cash transfer programs (e.g., Haryana’s Devirupak scheme), and incentives to

promote contraceptive prevalence (e.g., sterilization incentives in India). This paper examines

a novel policy experiment that imposes fertility limits on political candidates running for

leadership positions in the local government. Specifically, we analyze the impact of state-level

legislations that disbar individuals with more than two children from contesting Panchayat

and municipal elections on constituents’ fertility-related outcomes in India.

The rationale behind these laws is that limiting fertility of elected representatives can

decrease fertility among their constituents through a “role-model” effect. In addition, these

limits directly incentivize individuals who aspire to run for local office in the future to have

fewer children. The effectiveness of this policy measure in decreasing fertility, thus, depends

on the extent to which individuals’ fertility behavior is influenced by role-models and on the

attainability of a local leadership position for an ordinary citizen.

Starting with Rajasthan in 1992, eleven Indian states have enacted two-child limits on

local elected representatives. While four states revoked these laws after a few years of im-

plementation, it remains in effect in seven states. In all cases, the law included a one year

post-announcement grace-period, births during which were not counted towards the limit.

Our identification strategy exploits the quasi-experimental variation in the state and the year

of announcement of these laws to estimate their causal impact on fertility-related outcomes

in a differences-in-differences framework. We combine complete retrospective birth histories

from large-scale household surveys to construct a woman-year panel that spans the years

between 1973 to 2006. Broadly, we find that two-child limits on electoral candidates decrease

fertility in the general population but have unintended consequences for the sex ratio at

birth when son preference is strong.

1

Overall, the fertility limits significantly decrease the likelihood that a woman has more

than two children in a given year. Women who gave birth to two children before the two-

child limit was announced in their state are significantly less likely to have a third birth.

Consistent with the one-year grace-period, event-study analysis reveals that the likelihood of

third births does not change significantly in the first year after announcement, but declines

sharply thereafter. In addition, we examine how these effects vary across treatment states

by generating state-wise regression discontinuity (RD) type graphs.

We also examine the effect on the likelihood of second births and their sex ratio for women

whose first child was born before the laws were announced. We conduct this analysis in a

triple differences-in-differences setting by utilizing the variation in the sex of the firstborn

child. Prior literature shows that the sex of first births in India is close to random, despite

availability of prenatal sex-determination technology. We find that, due to the two-child

limits, women with firstborn girls are less likely to have a second birth and are more likely to

have a male child if there is a second birth, but only in upper-caste households. The negative

effect on second births can be explained by increased sex-selection as each abortion delays

the next birth, at the minimum, by a year (Bhalotra and Cochrane (2010)). These results

indicate that, among those whose first child was born before the treatment year, households

wishing to remain eligible for village council leadership restrict their fertility to two children,

but ensure that the second child is male if the first is not. Thus, a preference for sons can

cause policies that target lower fertility to have unintended effects on sex ratios.

We conduct a number of robustness checks to establish the causality of our findings. To

the best of our knowledge, these two-child limits in India are the first instance of a democratic

country instituting a fertility ceiling policy for candidates aspiring to elected political office.

This paper is, hence, the first to examine how households trade-off the chance to hold political

office against the option of having more children. We also investigate the role of son preference

in this trade-off by examining how fertility and sex selection responses to the policy differ

based on the number of sons when it is implemented.

2

Our paper makes novel contributions to two distinct literatures: on the effects of leaders’

characteristics on followers’ behaviors and on determinants of fertility and sex ratios in high-

son preference countries. Apart from the large literature on peer-effects, the socioeconomic

characteristics of individuals in positions of authority (e.g., teachers, mothers, and religious

leaders) have been shown to exert considerable influence on their followers’ behaviors and

outcomes (Fernandez et al. (2004), Bettinger and Long (2005), Olivetti et al. (2013), Bassi

and Rasul (2014)). Beaman et al. (2012) find that greater exposure to female leaders reduces

gender gaps in aspirations and educational attainment through a role-model effect. In a

related set of papers, exposure to television and specific social content has been shown to

affect viewers’ fertility rates (e.g., Jensen and Oster (2009), Chong et al. (2012)). Our novel

contribution is to show that fertility restrictions on local leaders affect their constituents’

demographic outcomes.

The literature on the determinants of fertility in developing countries is vast. Our findings

show that role-models and incentives for political office also affect fertility and sex ratios.

Recent work has highlighted the causal relationship between fertility decline and rising sex

ratios in societies like India where sons are preferred (Ebenstein (2010), Anukriti (2014),

Jayachandran (2014)). We augment this literature by investigating a new source of fertility

decline that has an unintended effect on sex ratios, similar to programs like the One Child

Policy and Devirupak.

In recent policy and political discussions in India, similar limits have been proposed for

state legislative assembly members as well as members of the national Parliament. To the

extent that individuals might find it easier to aspire to becoming a local leader and the

socioeconomic characteristics, especially fertility, of local politicians might be more salient

due to greater visibility, policies that affect local officials might be more effective than limits

on leaders situated in state or national capitals. Moreover, fertility-related policies often

ignore son preference. Given that sex-selective abortions are illegal in India, a biased sex

ratio at birth among local leaders can also have a negative role-model effect, as shown by

3

our sex ratio results, highlighting the need for more careful policy design.

The remainder of the paper is organized as follows. Section 2 discusses the two-child

limits in detail. Sections 3 and 4 describe our data and empirical strategy. Section 5 presents

the results and Section 7 concludes.

2 BackgroundIndia is the world’s second most populous country and houses a third of the world’s poorest

1.2 billion citizens (Olinto et al. (2013)). Consequently, fertility reduction continues to be

atop its policy agenda. Based on the recommendations of the Committee on Population

set up by the National Development Council (NDC) in 1992, several Indian states have

enacted legislations that disbar individuals with more than two children from contesting local

elections. The rationale behind these laws is that two-child norms for elected representatives

will decrease fertility among their constituents through a role-model effect. In addition, they

incentivize individuals who intend to run for elections in the future to plan smaller families.

In most states, the two-child limits have been enacted for elections to rural Panchayats,

however, a few states have also imposed these norms on urban municipalities. India has a

three-tiered decentralized system of local governance in rural areas, known as the Panchayati

Raj. It comprises village-level councils (Gram Panchayat), block-level councils (Panchayat

Samiti), and district-level councils (Zila Parishad). Although the Panchayat system has

existed in several Indian states since the 1950s, it was granted constitutional status in 1992

through the 73rd Amendment of the Indian Constitution (The Panchayati Raj (PR) Act).

Since then, regular Panchayat elections have taken place in most states. These elected local

councils receive funds from the national and the state governments and are authorized to

plan and implement developmental schemes as well as to levy and collect taxes. The Act

requires that at least one-third of all member and chief positions are reserved for women.

Similarly, positions are reserved for Scheduled Castes (SC) and Scheduled Tribes (ST) in

proportion to their share in the village, block, or district population. Reservations for these

4

groups are implemented in a stratified manner—among positions or seats reserved for SC,

ST, and “general” castes, one-third are randomly chosen for women.

Table 1 presents the timeline for the enactment and implementation of the two-child

laws across Indian states1 and Table A.1 shows the local election years for which they were

effective. Rajasthan was the first state to introduce such a law in 19922 and provided for

a one year grace-period—any births during the grace-period were not counted towards the

two-child limit. A candidate who had two or more children at the start of the cut-off and had

an additional child after the grace-period cut-off date was disqualified. However, no elections

were held under this amended law. The two-child norm was then included in Rajasthan’s

1994 PR Act which stipulated that anyone who had a third or higher-order birth after April

1994 would be ineligible to contest elections. Due to popular pressure, a grace-period was

provided whereby births during April 23, 1994 - November 27, 1995 were not counted towards

the two-child cut-off. As a result, the law came into effect after Rajasthan’s first post-73rd

Amendment Panchayat election (that took place in 1995). A similar law was also passed for

municipal elections in urban areas.

In Haryana, the law was announced through the PR Act in 1994 with a one-year grace

period (until April 24, 1995). However, the first Panchayat elections had already taken place

in 1994 and since members of the local councils are elected for a period of five years, no one

was disqualified during 1995-2000. The Haryana government revoked this law in July 2006

and the repeal came into effect retroactively from January 1, 2005.

Andhra Pradesh (AP) introduced the fertility limit in its 1994 PR Act and also provided

a one-year grace period. Orissa announced the law first for its district councils in November

1993 and then for the village and block councils in April 1994.3 Himachal Pradesh (HP),

1This information is largely based on Buch (2005) and Buch (2006).2Rajasthan’s law predates the NDC recommendations.3Additionally, in Orissa, an individual who cannot read and write Oriya or who has more than one living

spouse is also disqualified. The illiteracy criterion is not applicable to the village council elections.

5

Madhya Pradesh (MP), and Chhattisgarh4 introduced their laws in 2000 and, like Haryana,

repealed them in 2005. Maharashtra adopted the norm in 2003 with retrospective effect from

September 21, 2002. Lastly, Bihar and Uttarakhand have adopted the law for municipal

elections, but not for Panchayat elections.

Although their formulation is quite similar across states, the two-child laws are ambiguous

in some cases. For example, the laws in Haryana and MP explicitly mention two living

children, whereas in AP, Orissa, and Rajasthan the clauses do not distinguish between births

and living children. In Rajasthan, twins are considered as one birth and a still-birth is not

counted as a birth, while in MP the District Collector has discretion over disqualification

in these events. However, children given up for adoption are counted towards the two-child

limit for disqualification in all states. In most states, for a disqualification, a complaint has

to be filed with the appropriate adjudicating authority, except in Orissa (for village councils)

and MP where the competent authority can initiate action on its own.

Thus, starting in 1992, eleven Indian states have imposed a two-child limit on their local

politicians for at least a few years—four states revoked them in 2005,5 but they remain in

effect in seven major states. In all cases, a one year grace-period was provided, i.e., births

during one year after the announcement of the law were not counted towards the limit.

3 DataWe utilize repeated cross-sectional data from three rounds of the National Family Health

Survey (NFHS-1, 2, 3) and one round of the District-Level Household Survey (DLHS-2)

of India.6 Each survey-round is representative at the state-level and includes a complete

retrospective birth history for every woman interviewed, containing information on the month

4Chhattisgarh inherited the law when it was carved out of MP in 2000. Since 2004, candidates below 30years of age in Chhattisgarh are also required to be literate.

5The central government and the Union Ministry of Panchayati Raj encouraged these revocations (http://policydialogue.org/files/events/Aiyar_Key_Role_of_Panchayati_Raj_in_India.pdf).

6The years of survey are 1992-93, 1998-99, and 2005-06 for the NFHS and 2002-04 for the DLHS.

6

and the year of child’s birth, birth order, and mother’s age at birth. We combine these birth

histories to construct an unbalanced woman-year panel;7 a woman enters the panel in her

year of first marriage and exits in her year of survey.

For consistency across rounds, we limit the sample to currently-married women in the

15-44 age-group at the time of survey.8 We also drop women (i) who were married more

than 20 years prior to the survey to avoid issues related to imperfect recall, (ii) whose

husband’s age was below 15 or above 80 in the year of survey, and (iii) who have given birth

to more than ten children, to prevent any composition-bias since these women are likely to

be fundamentally different from rest of the sample. Lastly, we exclude mothers who have had

twins since multiple births in our context are largely unplanned and do not reflect parents’

fertility preferences.9 However, our results are not driven by any of these selection criteria.

Our final sample comprises 511,542 women and 1,261,711 births from 18 major states10

and covers the time period 1973-2006. As discussed earlier, the two-child laws were an-

nounced, enacted, and became effective (during an election) over several years. Moreover, a

one-year grace-period was provided in all instances. To err on the side of caution, we define

treatment based on the year of announcement, i.e., the earliest and the most conservative

year when the law might have had an effect. Since the most recent year in our sample is

2006, we cannot credibly examine the effect of revocations that took place in 2005. However,

we have a large number of post-announcement years, ranging from 4 to 13 years, to estimate

the relatively long-term effect of the fertility limits.

7The DLHS and the NFHS are similar in terms of the selection of respondents, the conduct of interviews,and the questionnaires used. Sample sizes, however, are much larger for the DLHS since it is also represen-tative at the district-level. As shown in Section 6, our results do not change if only one of these datasets isused.

8Survey questionnaires were administered to 13-49-year old ever-married women in NFHS-1, 15-49-yearold ever-married women in NFHS-2,3, and 15-44-year old currently-married women in DLHS-2.

9Additionally, we drop women who were visiting the household when the survey took place, and wereinterviewed as a result, since there is no information on their actual state of residence.

10The states of Uttarakhand, Jharkhand, and Chhattisgarh were, respectively, carved out from UttarPradesh (UP), Bihar, and MP in 2000. Since our data does not include districts-identifiers for all rounds, wesubsume these three new states into their parent states for our analyses.

7

Table 3 displays the years we use for defining the pre- and post-treatment period for

each affected state. Table 4 presents the sample means and standard deviations for the key

variables used in our analyses, separately for never-treated and treated states. We further

split the treated sample into pre- and post-treatment observations. About two-thirds of

women in our sample live in a rural area. A majority of them are Hindus, with a larger

share (90%) among treated relative to never-treated households (79%). In terms of caste-

composition, upper-castes and other backward classes (OBC) comprise about 40% and 35%

of the sample, while the rest belong to Scheduled Castes (SC) and Scheduled Tribes (ST).

Educational attainment is low for women, with more than half the sample being uneducated;

in comparison, 29% of the husbands are uneducated. In terms of our outcome variables,

women in the post-treatment group are less likely to give birth and are more likely to have

two children relative to women in the never-treated and pre-treatment sub-samples.

The sample means for the three groups in Table 3 are similar along most, if not all,

socio-economic dimensions. Nevertheless, to ensure that our estimates are not confounded

by any underlying differences between these samples, we control for religion, caste, standard

of living, husband’s and wife’s years of schooling, and residence in an urban area in our

regressions. To take into account state-specific factors, we include state fixed effects and also

control for state-specific linear time trends. In addition, we conduct several robustness checks

to establish that our estimates are measuring the causal effect of fertility limits.

4 Empirical StrategyThe goal of our empirical analyses is to estimate the causal effect of the two-child limits

imposed on local politicians in a state on fertility-related outcomes among residents in the

same state. To do so, we utilize the quasi-experimental geographical and temporal variation

in the announcement of these laws across Indian states. Although eleven states have enacted

such a law thus far, due to data limitations we can estimate its impact for only eight states:

Rajasthan, Haryana, AP, Orissa, HP, MP, Chhattisgarh, and Maharashtra. The law came

8

into effect in Bihar and Gujarat after 2006, so in our sample these states are not treated.

Although Uttarakhand announced its law for urban municipal elections in 2002, our analysis

excludes it from the group of treatment states because Uttarakhand was a part of Uttar

Pradesh until 2000 and we cannot distinguish between the two in the pre-2000 sample.11

Our results, however, are robust to the exclusion of Uttar Pradesh. In addition to Bihar,

Gujarat, and Uttarakhand, our control group comprises nine other states. Figure 1 depicts

the treatment and control states in a map.

If the two-child laws are effective, we expect to observe changes in the probability of

third parity births for couples who already have two children when the law is announced. To

examine if this is the case, we estimate the following differences-in-differences type regression

specification for a woman i of age a in state s and year t:

Yisat = α + β1Treatst +X′

iδ + γs + θt + ψa + νs ∗ t+ εisat (1)

where Treatst is equal to one for women residing in the treated states if t > the year

of announcement, and zero otherwise; γs, θt, and ψa are fixed-effects for state, year, and

woman’s age. We also control for state-specific linear time trends (νs ∗ t) and the following

covariates (Xi): five categories each for a woman’s and her husband’s years of schooling,

indicators for the religion (five categories), caste (four categories), and the standard of living

(three categories) of the household, residence in an urban area, and indicators for the year

of interview. We restrict the sample to women whose first two children are born before the

treatment is announced in their state. The key coefficient of interest is β1, which measures

the effect of two-child limits on our outcomes variable which is an indicator for a third birth.

It is likely that the two-child laws also affect second parity births for couples who have

one child at announcement. For example, if son preference is strong, women who have one

daughter when the law is announced may be more likely to practice sex-selection at second

11Note that Uttar Pradesh has never enacted a two-child limit for its local politicians.

9

parity due to the two-child limit. Therefore, we also estimate a triple differences-in-differences

version of (1) by interacting Treatst with an indicator (Girli) for whether the first child (born

before treatment) is a girl:

Yisat = α + β2Treatst ∗Girli + φTreatst + ωGirli

+X′

iδ + γs + θt + ψa + νs ∗ t+ τs ∗Girli + εisat

(2)

The outcome variables are indicators for a second birth and, conditional on birth, the

likelihood that the child is male. The coefficient φ estimates the effect of the two-child laws

on couples whose firstborn is a boy while β2 estimates the differential effect on couples

whose firstborn is a girl. Prior literature on India has shown that, despite the availability of

prenatal sex-determination technology, sex of the first birth is plausibly random (Bhalotra

and Cochrane (2010)) and most instances of sex-selection occur for higher-order births.

However, Anukriti (2014) finds that this is not true for first births in Haryana after 2002 when

firstborn children are more likely to be male due to the Devirupak scheme. Therefore, we

drop post-2002 observations for Haryana from our sample while estimating (2). In addition,

we restrict the sample to women whose first child is born before the year of treatment.

The inclusion of state and year fixed effects controls for any time-invariant state-level

variables and state-invariant overall time trends that might affect fertility outcomes. More-

over, state-specific time trends account for differential linear trends in fertility patterns across

states over the time period of analysis. We cluster standard errors at the state level when

both treated and never-treated states are included in the sample. In specifications where the

sample is restricted to only the treated states, we cluster at the state-year level to avoid

econometric issues pertaining to a small number of clusters.

Our underlying identifying assumption is that the state-year variation in the timing of

law announcement is uncorrelated with other time-varying determinants of the outcomes of

interest. In addition to controlling for state-specific linear trends in our regressions, in the

next section we show that there are no statistically significant differences in pre-treatment

10

trends for our treatment and control groups. This supports our identifying assumption that

the treatment and comparison women would have had similar trends in fertility rates in

the absence of the two-child limits. Moreover, we show that the timing of announcement is

uncorrelated with other socio-economic characteristics that vary by state and time.

5 Results

5.1 Graphical Evidence

Before discussing the regression results, we first present some graphical evidence for the

effect of the two-child limits. In Figure 2 we use the event-study framework to depict the

evolution of the likelihood that a woman has more than two living children in a given year.

The plotted coefficients show the differential trends in the likelihood of having more than

two living children for women in treatment and control groups, after controlling for socio-

economic characteristics of the woman and fixed effects for state, year, and woman’s age.12

There are no noticeable trends in the differential likelihood of having more than two

children in the pre-treatment years. This lack of significant differences in the years prior to

the two-child limits provides an important test for the validity of our identifying assumption;

the trends in outcomes across comparison groups evolve smoothly except through the change

in incentives for births in the treatment year. After the two-child limits are announced, there

is a sharp decrease in the probability that a woman reports having more than two living

children. However, this decrease is not immediate. The delayed effect can be explained by

the grace-period provision in the two-child laws that does not count births during one year

after the announcement of the law towards the fertility limit.

We also examine the effects of the law separately for each treatment state by plotting

the smoothed values of our outcome variable for years before and after the announcement.

12Specifically, Figure 2 plots the βk coefficients from the following regression: Yisat =∑5k=−10 βkTreats,t+k +X

′

iδ+γs +θt +ψa +εisat, where Treats,t+k indicates k years from the announcementof the law in state s. The year of announcement is the omitted year.

11

Specifically, we perform kernel-weighted local polynomial smoothing using the Epanechnikov

kernel and a bandwidth of 0.5. In Figures 3 and 4, the dependent variable is an indicator for

a third or higher parity birth. AP, Haryana, Rajasthan, and Orissa show strong increasing

trends in the likelihood of families having a third parity birth in the years leading up to the

two-child law. In all four states, however, this trend reverses after the law is announced and

the likelihood of third parity births declines sharply. MP shows no such trend before the law,

but also shows a sharp decline in third parity births once it is announced. The effects are

weaker for HP and Maharashtra, perhaps because the law was announced more recently in

these states and they already displayed a declining trend in third births preceding the law.

5.2 Regression Results

In this section we present regression estimates for the causal effects of the two-child limits

on (i) third parity births for women whose first two children were born before the laws were

announced, and (ii) second births for women whose first child was born before the laws were

announced.

5.2.1 Third births

Table 5 present estimation results for specification (1) to describe the effects of the two-

child limits on the likelihood of a third birth. We restrict the sample to women whose first

two children were born before the law was announced and to years after the second birth.

Column (1) controls for state and year fixed effects. In Column (2), we include additional

covariates that comprise indicators for the year of survey, woman’s age, household’s religion,

caste, wealth, husband’s and wife’s years of schooling, and residence in an urban area, and

state-specific linear time trends. Columns (3) and (4) restrict the sample to the treated

states. In Columns (5)-(7), we use observations from all states but restrict the sample by the

sex-composition of the first two births. The standard errors are clustered at the state-level

except in Column (4) where the sample is restricted to the treated states and hence we

cluster at the state-year level to avoid inference issues due to the small number of clusters.

12

The coefficients for Treatst are negative and significant in the first two columns implying

that the two-child limits decreased higher-order fertility for couples who had already borne

two children by the time the law was announced in their state. The coefficient in Column

(2) translates into a 0.5 percentage point or a 5.5 percent decrease in the likelihood of a

third birth from the baseline probability of 9 percent. The coefficients remain negative and

significant when we restrict the sample to the treated states in Columns (3) and (4).

Next we examine if these effects vary by the sex-composition of pre-existing children when

treatment began. To the extent that son preference is quite strong in some of the treated

states, we expect couples who have one or two sons prior to the treatment to be more likely

to stop childbearing relative to those without any sons. On the other hand, as reflected by

the control group means, these women are more likely (than those without any sons) to stop

childbearing even in the absence of the two-child limits and hence the incremental effect of

the two-child limits may not be large. In addition, couples who have two female births to

begin with perhaps have a weaker preference for sons and hence may be more likely to reduce

fertility at the margin. Although all three coefficients in Columns (5)-(7) are negative, we do

in fact find that the decrease in marginal fertility is significant only for women whose first

two births were female.

5.2.2 Second births

In Table 6, we present results for the probability and sex of the second birth by sex of

the first child. We restrict the sample to women whose first child was born before the law

was announced and to years after the first birth. Column (1) shows that before the law is

announced, a firstborn girl, relative to a firstborn boy, increases the probability of a second

birth by 0.2 percentage points, reflecting parents’ desire for at least one son. However, once

the law is announced, there is a decrease in the likelihood of a second birth, with a larger

decrease for those with a firstborn girl. Splitting the sample by the caste of the household in

Columns (2) and (3) reveals that this effect is primarily driven by upper-caste families with a

firstborn girl. While this decrease could imply a permanent decrease in fertility, it could also

13

reflect a relative delay in second births due to greater sex-selection by women with firstborn

girls. If the latter is true, we expect the sex ratio of second births to increase for families with

a stronger preference for sons and a greater likelihood of contesting local elections. Thus,

in Columns (4)-(6), we examine the effect of the two-child limits on the likelihood that the

second child is male.

While the coefficients in the first two rows of Column (4) are positive, there is no signif-

icant effect of the two-child laws on sex-selection behavior in the overall sample. However,

consistent with Columns (2) and (3), disaggregating the sample by caste reveals that the

limits increase sex-selection for second births by 3 percentage points among upper-caste fam-

ilies if their first child is a girl. Lower-caste households, on the other hand, do not respond

in the same manner, despite being 1.23 percentage points likelier to have male second-born

children if their first child is a girl (relative to a boy) before the law is announced. This

pattern of results suggests that the decrease in second parity births reflects a delay induced

by greater sex-selection. Due to their dominant socioeconomic status, we expect upper-caste

households to be more concerned about remaining eligible for village council leadership than

lower-caste households. Moreover, prior literature suggests that they also have a stronger

preference for sons. Consequently, if their first child is a girl, they increase sex-selection at

second births to ensure that they at least have one son whilst not sacrificing future eligibility

for political office.

6 RobustnessIn this section we perform a few robustness checks to ensure that our previous results truly

capture the causal effect of fertility limits on politicians. First, we conduct a placebo test

by reassigning the intervention or treatment to a year before the actual law was announced.

If our results are truly capturing the causal effect of the two-child laws, we should not

find significant effects in these placebo regressions. Table 7 presents the results from these

regressions. Each column uses a different year as a placebo treatment year. For example,

14

in Column (1), we assume that the two-child laws were announced in all treatment states

in 1980. Since these laws are fictitious, a significant “effect” at the 5% level may be found

roughly 5% of the time. There is no cell where we find a significant effect in the same direction

as our main results in Table 5. These findings lend support to our differences-in-differences

estimation strategy and make a causal interpretation more credible.

Next we examine the effect of these laws on the likelihood of fourth births for couples

whose first three children were born before the announcements. If our findings are indeed

causal, we should not find a significant effect since the two-child limits are irrelevant for

couples who had already given birth to three children prior to the treatment. The coefficient

in Column (1) of Table 8 is insignificant, further suggesting that our results are not driven

by a general decline in marginal births for all parities. Column (2) shows that our results

also remain robust when only NFHS data is used, thereby addressing concerns about the

bias introduced by any unobserved differences in data collection, or small variations in the

sampling methodology for NFHS and DLHS.

One potential mechanism through which these laws can affect fertility outcomes is through

adjustments in the age at marriage. Forward-looking individuals (or their parents) wishing

to maintain eligibility for local elections in the future may take into account the lower com-

pleted fertility requirements (i.e., a maximum of two children) and delay marriage, which

could explain the decrease in likelihood of birth we observe in Section 5. To test if this is

the case, we estimate specification (1) with age at first marriage as the dependent variable.

The results are presented in Column (3) of Table 8 and show that there is no impact of the

two-child limits on the age at first marriage.

Although we control for a number of socio-economic variables in our regressions, to further

support our identification strategy, we show that the timing of announcement of these laws

across states is uncorrelated with changes in these characteristics that vary across states and

over time. Specifically, in Table 9 we present the coefficients from regressions that use various

maternal, paternal, and household characteristics as dependent variables in the estimation of

15

equation (1) with state and year fixed effects, and state-specific time trends, but without any

other controls. Out of 20 coefficients, the only marginally significant coefficient is a negative

effect on the likelihood of the mother being Hindu.

7 ConclusionThis paper examines whether demographic characteristics of locally elected representa-

tives affect their constituents’ fertility and sex ratio outcomes. To do so, we utilize quasi-

experimental variation in the enactment of two-child eligibility requirements for individuals

running for office in India. Our results show that fertility limits on local officials successfully

lower fertility among the general populace, but also lead to an unintended increase in the

already male-biased sex ratio in certain socioeconomic groups. We highlight a new channel

of demographic influence, namely local politicians.

Our results thus far can be explained by the role-model effect as well as the incentive

effect for individuals aspiring to run for office in the future. The data from NFHS and DLHS

do not allow us to distinguish between these two channels. In ongoing work, we seek to

exploit variation in the gender- and caste-based reservation status of village councils as an

exogenous shock to the “attainability” of these leadership positions. Moreover, we plan to use

data from the Rural Economic and Demographic Survey (REDS) to examine heterogeneity

in the effects of these laws by the presence of a family member who has contested or been

elected to a local council in the past.

Lastly, to the extent that women and low-caste households might have relatively less

control over their fertility decisions, these laws may have unintended consequences for the

political representation of socioeconomically disadvantaged groups who have relatively higher

fertility. Therefore, in future work, we would also like to examine the interactions between

these fertility limits and caste- and gender-based reservations in terms of their effects on the

characteristics of elected candidates.

16

ReferencesAnukriti, S. (2014): “The Fertility-Sex Ratio Trade-off: Unintended Consequences of Fi-nancial Incentives,” IZA Discussion Paper No. 8044.

Bassi, V. and I. Rasul (2014): “Persuasion: A Case Study of Papal Influence on FertilityPreferences and Behavior,” .

Beaman, L., E. Duflo, R. Pande, and P. Topalova (2012): “Female LeadershipRaises Aspirations and Educational Attainment for Girls: A Policy Experiment in India,”Science, 335, 581–586.

Bettinger, E. P. and B. T. Long (2005): “Do Faculty Serve as Role Models? TheImpact of Instructor Gender on Female Students Do Faculty Serve as Role Models? TheImpact of Instructor Gender on Female Students,” American Economic Review, 95, 152–7.

Bhalotra, S. and T. Cochrane (2010): “Where Have All the Young Girls Gone? Iden-tification of Sex Selection in India,” IZA Discussion Paper No. 5381.

Buch, N. (2005): “Law of Two-Child Norm in Panchayats: Implications, Consequences andExperiences,” Economic and Political Weekly, XL.

——— (2006): The Law of Two Child Norm in Panchayats, Concept Publishing Company.

Chong, A., S. Duryea, and E. La Ferrara (2012): “Soap Operas and Fertility: Evi-dence from Brazil,” American Economic Journal: Applied Economics, 4, 1–31.

Ebenstein, A. (2010): “The “Missing" Girls of China and the Unintended Consequences ofthe One Child Policy,” Journal of Human Resources, 45, 87–115.

Fernandez, R., A. Fogli, and C. Olivetti (2004): “Mothers and Sons: PreferenceFormation and Female Labor Force Dynamics,” Quarterly Journal of Economics, 119,1249–99.

Jayachandran, S. (2014): “Fertility Decline and Missing Women,” NBER Working Paper20272.

Jensen, R. and E. Oster (2009): “The Power of TV: Cable Television and Women’sStatus in India,” Quarterly Journal of Economics, 124, 1057–94.

Olinto, P., K. Beegle, C. Sobrado, and H. Uematsu (2013): “The State of the Poor:Where are the Poor, Where is Extreme Poverty Harder to End, and What is the CurrentProfile of the World’s Poor?” Economic Premise.

Olivetti, C., E. Patacchini, and Y. Zenou (2013): “Mothers, Friends and GenderIdentity,” NBER Working Paper 19610.

Visaria, L., A. Acharya, and F. Raj (2006): “Two-Child Norm: Victimising the Vul-nerable?” Economic and Political Weekly, XLI.

17

8 Figures

Figure 1: Treatment and Control States

18

Figure 2: Likelihood of more than two living children, by year

NOTES: This figure plots the βk coefficients and their 95% confidence intervals (dashed lines) from estimatingthe following equation:Yisat =

∑4k=−10 βkTreats,t+k + X

′

iδ + γs + θt + ψa + εisat, where Treats,t+k indicates k years from theannouncement of the law in state s. Standard errors are clustered by state-year. The first vertical line (atk = 0) indicates the year of announcement. The second vertical line indicates the end of the one-year graceperiod. The sample is restricted to women in treatment states.

19

Figure 3: Likelihood of third or higher order births, by state

NOTES: This figure plots the smoothed values of the outcome variable (an indicator for third or higherorder birth) for years before and after the announcement of the law using kernel-weighted local polynomialsmoothing with an Epanechnikov kernel and a bandwidth of 0.5. The dashed lines indicate the 95% confidenceintervals.

20

Figure 4: Likelihood of third or higher order births, by state

NOTES: This figure plots the smoothed values of the outcome variable (an indicator for third or higherorder birth) for years before and after the announcement of the law using kernel-weighted local polynomialsmoothing with an Epanechnikov kernel and a bandwidth of 0.5. The dashed lines indicate the 95% confidenceintervals.

21

9 Tables

Table 1: Timeline for Two-Child Norms across States

State Announced Grace Period In effect End

Rajasthan 1992 Apr 23, 1994 - Nov 27, 1995 Nov 27, 1995 -Haryana 1994 Apr 21, 1994 - Apr 24, 1995 Apr 25, 1995 - Dec 31, 2004 Jul 21, 2006

(retro. impl. Jan 1, 2005)Andhra Pradesh 1994 May 30, 1994 - May 30, 1995 Jun 1995 -Orissa 1993/199413 Apr 1994 - Apr 21, 1995 Apr 22, 1995 -Himachal Pradesh 2000 Apr 18, 2000 - Apr 18, 2001 Apr 2001 - Apr 2005 Apr 5, 2005Madhya Pradesh 200014 Mar 29, 2000 - Jan 26, 2001 Jan 2001 - Nov 2005 Nov 20, 2005Chhattisgarh 2000 2000 - Jan 2001 Jan 2001- 2005 2005 (earliest mention)13

Maharashtra 200315 Sep 21, 2002 - Sep 20, 2003 Sep 2003 -Uttarakhand (municipal only) 2002Gujarat 2005 Aug 2005 - Aug 11, 2006 Aug 11, 2006 -Bihar (municipal only) Jan 2007 Feb 1, 2007 - Feb 1, 2008 Feb 1, 2008 -

13For district councils in 1993 and for village and block councils in 1994.14Notified on May 31, 2000. This created problems since people whose third child was born in Jan 2001 contested their disqualification for birth

within 8 months of the new law.15In retrospective effect from Sep 21, 2002.

22

Table 2: Panchayat members disqualified during 2000-04, selected states

State Number of disqualifications

Haryana 1,342Rajasthan 548Madhya Pradesh 862Andhra Pradesh 94*

SOURCE: Visaria et al. (2006). *Data available for 15 out of 23 districts.

Table 3: Treatment years, by state

State Treatst = 1 if year >

Rajasthan 1993Orissa 1993Haryana 1994Andhra Pradesh 1994Himachal Pradesh 2000Madhya Pradesh (inc. Chhattisgarh) 2000Maharashtra 2002

23

Table 4: Summary Statistics

Never treated TreatedPost = 0 Post = 1

Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.(1) (2) (3) (4) (5) (6)

Urban 0.343 0.475 0.329 0.470 0.320 0.466Hindu 0.786 0.410 0.897 0.304 0.898 0.303Muslim 0.161 0.367 0.066 0.249 0.063 0.243Sikh 0.041 0.198 0.010 0.100 0.013 0.113Christian 0.027 0.162 0.011 0.103 0.014 0.117SC 0.180 0.384 0.160 0.367 0.177 0.382ST 0.062 0.240 0.149 0.356 0.134 0.341OBC 0.365 0.481 0.298 0.457 0.374 0.484Wife’s years of schooling:Zero 0.514 0.500 0.563 0.496 0.544 0.4985-10 years 0.244 0.429 0.229 0.420 0.235 0.42410-12 years 0.091 0.287 0.074 0.261 0.082 0.27512-15 years 0.048 0.214 0.031 0.173 0.039 0.193≥ 15 years 0.045 0.207 0.037 0.188 0.046 0.209Husband’s years of schooling:Zero 0.278 0.448 0.291 0.454 0.289 0.4535-10 years 0.301 0.459 0.309 0.462 0.310 0.46210-12 years 0.153 0.360 0.149 0.357 0.149 0.35612-15 years 0.093 0.290 0.070 0.255 0.079 0.270≥ 15 years 0.096 0.294 0.089 0.285 0.101 0.302Low SLI 0.446 0.497 0.460 0.498 0.425 0.494High SLI 0.242 0.428 0.233 0.423 0.250 0.433Mother’s age at birth 24.853 6.163 23.008 5.474 26.507 6.341Birth = 1 0.213 0.410 0.239 0.426 0.161 0.367Birth is male 0.111 0.315 0.124 0.330 0.085 0.278Has 2 children 0.260 0.438 0.234 0.423 0.287 0.442

N 3,568,675 1,458,849 941,801

NOTES: Post is defined using the year of announcement of the law (see Table 3). SC, ST, and OBC indicateScheduled Caste, Scheduled Tribe, and Other Backward Class households, respectively. Low and High SLI areequal to one if the household belongs to the bottom-third or the top-third of household wealth distribution.

24

Table 5: Effects on Third Births

Only treated states BB BG GG

Dep Var: 3rd birth = 1 (1) (2) (3) (4) (5) (6) (7)

Treatst -0.0200*** -0.0049* -0.0068** -0.0068*** -0.0033 -0.0050 -0.0069**[0.0054] [0.0028] [0.0024] [0.0022] [0.0023] [0.0032] [0.0032]

N 2,899,022 2,880,757 1,059,213 1,059,213 773,470 1,442,666 664,621

Control group mean 0.080 0.080 0.098 0.098 0.072 0.078 0.091

Year FE x x x x x x xState FE x x x x x x xCovariates x x x x x xState-specific linear trends x x x x x xClustering State State State State-Year State State StateN (clusters) 18 18 18 224 18 18 18

NOTES: This table reports the coefficients of Treatst from specification (1). Each coefficient is from a separate regression. The dependent variable isone if there is a third birth in a given year, and zero otherwise. The sample is restricted to women whose first two children were born before the lawwas announced in her state. Only years after the second birth are included. Other covariates comprise indicators for the year of survey, woman’s age,household’s religion (Hindu, Muslim, Sikh, Christian), caste (SC, ST, OBC), wealth (low and high SLI), husband’s and wife’s years of schooling (5categories each), and residence in an urban area. In columns (3)-(4), the sample is restricted to women in treatment states. BB, BG, GG respectivelyindicate the sub-samples of women whose first two births were two boys, one boy-one girl, and two girls. *** 1%, ** 5%, * 10%.

25

Table 6: Effects on second birth, by first child’s sex

2nd birth = 1 2nd birth is male

Upper-caste Lower-caste Upper-caste Lower-caste

(1) (2) (3) (4) (5) (6)

Treatst * First− born girl -0.0029** -0.0030** -0.0026 0.0073 0.0307*** -0.0017[0.0012] [0.0013] [0.0015] [0.0064] [0.0093] [0.0061]

Treatst -0.0041* -0.0044 -0.0037 0.0039 -0.0065 0.0054[0.0023] [0.0028] [0.0025] [0.0050] [0.0082] [0.0051]

First− born girl 0.0024*** 0.0023*** 0.0023** 0.0103*** 0.0070*** 0.0123***[0.0007] [0.0006] [0.0009] [0.0013] [0.0016] [0.0014]

N 4,088,203 1,587,439 2,500,764 329,905 126,712 203,193

NOTES: The sample is restricted to women whose first child was born before the law was announced in her state. The dependent variable is one ifthere is a second birth in a given year, and zero otherwise. Only years after the first birth are included. Columns (4)-(6) are conditional on a secondbirth. We drop post-2002 observations for Haryana. Each coefficient is from a separate regression that includes state-specific linear time trends, fixedeffects for state, year, and the interaction between state indicators and first-born girl dummy. Standard errors are in brackets and are clustered bystate. Covariates comprise indicators for the year of survey, woman’s age, household’s religion (Hindu, Muslim, Sikh, Christian), (except in columns 3and 4) caste (SC, ST, OBC), wealth (low and high SLI), husband’s and wife’s years of schooling (5 categories each), and residence in an urban area.Lower-caste refers to SC, ST, OBC households; Upper-caste comprises the rest. *** 1%, ** 5%, * 10%.

26

Table 7: Placebo Test for Likelihood of Third Birth

Placebo treatment year:

↓ Dep var: 1982 1983 1984 1985 1986 1987 1988 1989Third birth =1 (1) (2) (3) (4) (5) (6) (7) (8)

Treatst 0.013 0.011* 0.006 0.006 0.005 0.001 -0.002 -0.002[0.007] [0.006] [0.005] [0.006] [0.005] [0.005] [0.004] [0.003]

N 2,880,757

NOTES: Each coefficient is from a separate regression with a different placebo treatment year (same for alltreated states). The dependent variable is one if there is a third birth in a given year, and zero otherwise.The sample is restricted to women whose first three children were born before the law was announced inher state. Only years after the third birth are included. Standard errors are in brackets and are clustered bystate. Specifications are similar to column (3) in Table 5. *** 1%, ** 5%, * 10%.

Table 8: Robustness Checks

Dep Var → 4th birth = 1 NFHS only Age at first marriage(1) (2) (3)

Treatst -0.0057 -0.006** 0.005

[0.0033] [0.002] [0.336]

N 1,631,630 876,382 62,401

Year FE x x xState FE x x xCovariates x x xState-specific linear trends x x x

NOTES: Each coefficient is from a separate regression. In Column (1), the specification is similar to that inTable 5. The dependent variable is one if there is a fourth birth in a given year, and zero otherwise. Thesample is restricted to women whose first three children were born before the law was announced in herstate. Only years after the third birth are included. In Column (2), the sample is restricted to NFHS data.In Column (3), the sample is restricted to one observation per woman. Treatst is equal to one if a woman’sfirst marriage took place after the law was announced in her state, and zero otherwise. Standard errors are inbrackets and are clustered by state. Other covariates comprise indicators for the year of survey, household’sreligion (Hindu, Muslim, Sikh, Christian), caste (SC, ST, OBC), wealth (low and high SLI), husband’s andwife’s years of schooling (5 categories each), and residence in an urban area. *** 1%, ** 5%, * 10%.

27

Table 9: Correlations between Law Announcements and Socioeconomic Variables

Coefficient of Treatst Std. ErrorDependent Variable (1) (2)

Urban 0.007 [0.009]SC -0.003 [0.002]ST 0.006 [0.005]OBC 0.007 [0.006]Hindu -0.005* [0.003]Muslim 0.001 [0.002]Sikh 0.0005 [0.001]Christian -0.001 [0.002]Low SLI -0.001 [0.005]High SLI 0.002 [0.005]Wife’s years of schooling:Zero -0.002 [0.003]5-10 years 0.001 [0.003]10-12 years 0.001 [0.002]12-15 years 0.002 [0.002]≥ 15 years -0.0001 [0.001]Husband’s years of schooling:Zero -0.001 [0.002]5-10 years 0.00009 [0.002]10-12 years 0.001 [0.002]12-15 years 0.002 [0.003]≥ 15 years -0.001 [0.002]

N 5,969,325

NOTES: Each coefficient is from a separate regression that includes state and year fixed effects and state-specific linear time trends. Standard errors are in brackets and are clustered by state. *** 1%, ** 5%, *10%.

28

A Appendix

Table A.1: Panchayat Elections

State Election YearsW/o the norm With the norm

Rajasthan 1995 2000, 2005, 2010Haryana 1994, 2010 2000, 2005 (?)Andhra Pradesh 1995, 2001, 2006, 2011Orissa 1997, 2002, 2007, 2012Himachal Pradesh 1995, 2005, 2010-11 2000Madhya Pradesh 1994, 2010 200016, 2005Chhattisgarh 2010 2000, 2005Maharashtra 1995, 2000 2007, 2010, 2013Uttarakhand 2003, 2008, 2014Jharkhand 2010Gujarat 2001, 2005-06 2010-11Bihar 2006 2011

16Despite the fact that the two-child norm was introduced after the panchayat elections were over in 2000,the new government started disqualifying elected representatives Visaria et al. (2006).

29

Figure A.1: Likelihood of third births, by year

NOTES: This figure plots the βk coefficients and their 95% confidence intervals (dashed lines) from estimatingthe following equation:Yisat =

∑4k=−10 βkTreats,t+k + X

′

iδ + γs + θt + ψa + εisat, where Treats,t+k indicates k years from theannouncement of the law in state s. Standard errors are clustered by state-year. The vertical line (at k = 0)indicates the year of announcement. The sample is restricted to women in treatment states whose first twochildren were born before the law was announced in her state. Only years after the second birth are included.

30